Silence Is Not Coverage

A Canadian practitioner asks whether using AI increases their professional exposure. The honest answer has two parts, and the second is the one nobody expects: the standard they are held to is unchanged, and the arrangements they rely on to survive a claim may not have been written with any of this in mind.

Scope And Sources

Two disclosures before anything else, because this article touches subjects where imprecision would be harmful.

This publication is an accounting and advisory practice. Nothing here is legal advice or insurance advice, and no reader should act on it without their own broker, insurer or counsel.

And the sources are overwhelmingly American and concern the legal profession. Insurance arrangements for Canadian professionals differ materially, including because several professions here operate under mandatory or profession-sponsored programmes rather than an open commercial market. We have flagged this throughout and devoted a section to what does and does not carry across.

The reason to report it anyway is that the mechanisms are structural rather than jurisdictional. A policy written before a technology existed will be interpreted against its existing language wherever it was issued, and the questions a Canadian practitioner should put to their own provider are the same questions regardless of who answers them.

The Standard Did Not Change

The part that is settled, and that this publication has stated consistently.

A professional who signs work is answerable for that work. The method used to prepare it is not a defence, and it never has been. A practitioner who relied on a junior's calculation, a template, a software output or a model is in the same position: they adopted the work by signing it.

Earlier articles in this series made the same point from other directions. The lineage article argued that a figure produced with AI assistance entering a financial statement or a client deliverable inherits the requirement to be traceable to its support, and that this predates all of the technology. The oversight article argued that a professional who signs is accountable regardless of what produced the draft, and that a review which did not function is not a defence.

So the standard of care question is, in itself, uninteresting. It is the same standard, applied to work prepared differently.

What has changed is everything downstream of that: whether the loss is characterised in a way an insurer will pay, whether a regulator takes a separate interest, and whether the practitioner can demonstrate what they actually did. Those are the subjects of this article.

Ambiguity, Not Coverage

The central finding, stated in one source in a sentence worth committing to memory.

One source observes that most 2022-era policies do not mention AI at all, which creates ambiguity, not coverage[1].

Another reports that a legal malpractice carrier has publicly stated that coverage for AI-related claims is uncertain under current policy language, and adds that the uncertainty benefits the insurer rather than the insured[2]. The same source states that most firms are using AI with professional liability insurance that was never designed to cover it[2].

The instinct a Canadian practitioner brings to this is that a policy which does not exclude something covers it. That instinct is not unreasonable and it is not how a contested claim works.

Coverage turns on whether the loss falls within the insuring clause, which typically responds to wrongful acts in the performance of professional services. Silence about a new method does not resolve whether the loss arose from professional services performed by the insured, and that is precisely the question the next section shows can be argued either way.

The practical consequence is that a firm cannot establish its position by reading its policy for the word AI and finding nothing. The absence is the problem rather than the reassurance.

The Recharacterisation Problem

The mechanism, and the most important thing in this article to understand.

One source sets out the fork: the question is whether AI-assisted work qualifies as professional services performed by the insured, and if the insurer argues the AI performed the work rather than the professional, they could deny coverage, while if the insurer argues the professional failed to supervise the AI, being a supervisory failure, coverage may apply but the claim analysis changes[2].

Another describes the same move from the other direction: a client is penalised, sues, and the malpractice insurer argues this was a technology failure rather than a professional judgment error, so the gap is that the insurer recharacterises the claim to avoid coverage[3]. It notes that policy exclusions for technology failures could be invoked to deny or limit coverage[1].

The structure is that one set of facts supports two characterisations. A wrong figure in a deliverable can be described as a professional who exercised judgment badly, or as a tool that produced a bad output which a professional passed along.

The first characterisation sits inside a professional liability policy. The second points toward technology exclusions, or toward a different policy entirely.

And the party selecting the characterisation in the first instance is the one deciding whether to pay. That is not an accusation of bad faith; it is how coverage disputes work, and it is why the recommendation below is to obtain a position in writing before there is a claim rather than after.

Neither Policy Clearly Responds

The consequence of the fork, which is worse than either branch alone.

One source describes the result: when an AI tool contributes to a malpractice event, the claim lands in a grey zone where neither the professional liability policy nor the cyber liability policy clearly responds[1].

A firm carrying both may reasonably assume that between them everything is covered. The two forms were drafted to cover different things and the boundary between them was drawn before this category of loss existed.

Professional liability responds to errors in professional judgment. Cyber responds to security incidents, data breaches and system compromise. An AI-produced error is neither a security incident nor, on one available characterisation, a judgment error.

The recommendation one source gives is precise and is the single most actionable item here: obtain written confirmation that AI verification failures are covered as professional services claims rather than as technology failures[3].

That framing, verification failures as professional services claims, is the right one to put to a Canadian provider, because it describes what actually happens: the professional reviewed the output and the review did not catch the defect.

Silent And Affirmative Cover

The vocabulary, which is useful for a conversation with a broker.

One source, which discloses that its carrier actions and form citations are verified against primary sources or named broker and trade-press confirmation and states plainly that it is not legal or insurance advice, describes two emerging concepts. Silent AI cover is where an existing professional liability or errors and omissions form has not been amended for AI, so the carrier adjudicates an AI-related claim under pre-existing policy language. The alternative is affirmative AI cover. It states that whether the professional liability policy actually pays an AI-driven claim is the load-bearing question, and that in 2026 the answer is bifurcated[4].

We credit that source's methodological disclosure, which is more careful than most of this literature.

The distinction gives a Canadian practitioner a precise question rather than a vague worry. Not "am I covered for AI," which invites a reassuring non-answer, but "is our cover silent or affirmative on AI-assisted work, and if silent, what is the carrier's position."

The word bifurcated is also worth noting. The market is not moving in one direction. Some carriers are adding exclusions and some are adding affirmative endorsements, which means the answer depends entirely on the provider and cannot be inferred from general commentary.

Exclusions Are Arriving

The market movement, reported with attention to source quality because the sources here differ in reliability.

A trade publication reports that one insurer has been leading on broad AI exclusions in its professional liability forms, introducing exclusions as far back as 2024, and that this emerging trend in the legal space could be a predecessor to more widespread AI exclusions in the professional lines market, with some carriers already implementing them across multiple industry verticals[5].

A career and jobs publication reports more broadly that several major carriers filed regulatory exclusions for AI-related claims in 2026, stripping the technology out of standard commercial liability and directors and officers policies, and that others have obtained regulator approval to attach AI exclusion clauses to commercial liability policies[6].

We report the second with more caution than the first. It is a general-interest publication rather than an insurance trade source, we have not verified the filings, and named carrier claims of this kind warrant checking before reliance.

A third source states that AI exclusion clauses are appearing in new renewals at an accelerating rate, with some carriers inserting blanket exclusions[1].

The transferable observation for a Canadian practitioner is about timing rather than about any named insurer. If exclusions are being introduced at renewal, then a firm's position can change on a date it already has in its diary, without negotiation and without the firm noticing, because renewal documents are not read the way new policies are.

The same sources report movement the other way, with one large carrier developing endorsement options and another offering an AI endorsement with explicit coverage for AI-assisted work[1], so the answer for any individual firm is a question rather than a forecast.

The Question On The Renewal Form

The mechanism by which this reaches a Canadian firm first, and a detail with unusual relevance to this publication's readers.

One source quotes a senior risk control lead at a major broker, whose remit is described as covering accounting and law firm professional services, summarising the position in April 2026 as carriers asking firms whether they use AI[4]. Another reports that a large carrier has introduced AI-specific renewal questionnaires[1].

The reference to accounting practice specifically matters, because most of this commentary concerns lawyers and this is one of the few points where the accounting profession is named.

A renewal questionnaire is not merely information gathering. It is the point at which a carrier begins pricing and underwriting the exposure, and the answers become part of the record on which cover is issued.

For a Canadian firm the practical consequence is that this subject arrives whether or not the firm has thought about it, on the renewal date, in the form of a question that has to be answered accurately.

Answering It Accurately Is Harder Than It Sounds

A point we have not seen made in this literature and which we think is the most immediately useful thing here. This section is our own analysis.

Consider a Canadian practice asked at renewal whether it uses AI. The partner completing the form thinks about whether the firm has deployed an AI tool, concludes that it has not, and answers accordingly.

Meanwhile the accounting platform has AI-assisted categorisation switched on. The document management system summarises files. Staff use a general-purpose assistant on their own accounts for drafting. Someone built a workflow with a spreadsheet add-in.

The answer given was honest and may not be accurate, and inaccuracy in a proposal or renewal form is a category of problem entirely separate from whether the underlying risk is covered.

We are not going to state the legal effect of a misstatement in an insurance application, which varies and is a matter for counsel. We will say that a firm which cannot enumerate where AI is in use is not in a position to answer the question at all, and that most Canadian firms cannot.

That produces a concrete recommendation with a deadline attached to it: build an inventory of AI use before your next renewal date, covering deployed tools, features enabled inside existing software, and what staff use on their own initiative. The last category is the one that will be missing and the one most likely to surprise.

The Exposure Insurance Does Not Reach

The point most relevant to a licensed Canadian professional, and one this literature mentions only in passing.

One source notes that workers in licensed professions including medicine, law and accounting face an additional layer, because their professional licensing body can sanction them for AI-assisted mistakes regardless of who is sued[6].

That sentence identifies a structurally different exposure, and this is our own elaboration.

Professional liability insurance responds to claims: a third party alleges loss and seeks money. Regulatory discipline is not a claim. It is a proceeding by the profession about whether a member met its standards, and its outcomes are reprimand, conditions, suspension or removal from practice.

The two are independent. A practitioner can face discipline with no claim, because no client lost money but the conduct fell below standard. A practitioner can settle a claim entirely within cover and still face a separate regulatory process about the same facts.

And discipline is uninsurable in the sense that matters: cover may fund a defence, and it cannot restore a licence.

For a Canadian CPA or other regulated professional, this means the insurance conversation is necessary and insufficient. The question that outranks it is whether the work meets the profession's standard, which is a question about the firm's own practices rather than about anyone's policy wording.

Competence Now Includes The Technology

The conduct rules position, reported from a foreign framework with its Canadian analogue flagged.

One source describes the American baseline as competence, diligence and supervision of non-professional assistance, and notes that a comment to the competence rule explicitly states that competence includes keeping abreast of changes in the law and the benefits and risks associated with relevant technology, with a reported forty states having adopted some version of that technology competence language by 2023[7].

We report that as a description of a foreign regime and have not verified the state adoption count.

The structural idea is what transfers. A duty of competence framed as including relevant technology means that not understanding the tool is itself a potential failure, rather than an excuse for a failure.

That inverts how many practitioners think about this. The instinct is that unfamiliarity with a system mitigates responsibility for its output. Under a technology competence framing it does the opposite.

Canadian professional bodies have their own codes and their own guidance, and members should work from those rather than from an American commentary. The observation worth carrying into that reading is that the relevant question may not be only whether the work was checked, but whether the practitioner understood the failure modes of what they were checking. This series has spent eighteen articles on exactly those failure modes.

Where Exposure Attaches

A useful decomposition from the same source.

It states that malpractice exposure attaches differently depending on which stage of the work AI touches, how much supervision occurred, and whether the output was verified before reliance[7].

Those three variables map onto arguments made earlier in this series.

Which stage. The function allocation article argued that automation level should be chosen per stage rather than per task, and that the execution stage should be governed by reversibility. A firm that automated gathering and analysis while retaining the deciding stage is in a materially different position from one that automated the conclusion.

How much supervision. The oversight article argued that most review designs are post-hoc and supervisory, that explanations increase deference, and that constitutive designs where the human sets criteria in advance are stronger.

Whether verified before reliance. The offloading article found that checking claims against sources predicted better outcomes, and that verification should be scheduled rather than triggered by felt uncertainty because felt uncertainty is an unreliable trigger.

The point is that the exposure analysis and the design analysis are the same analysis. A workflow designed well on the grounds this series has argued is also a workflow that is easier to defend, and that is not a coincidence: both ask what the human actually did.

An Uncomfortable Consequence Of The Research

An implication of this series' own findings that cuts against practitioners, and which we would rather state than leave for someone else to notice. This section is our own analysis.

The automation bias article reported peer-reviewed research establishing that complacency and bias occur in expert participants, cannot be prevented by training, and arise from attentional conditions rather than individual diligence. We used that to argue that blaming a reviewer misidentifies the cause.

Now consider the same research in a liability setting.

In the coverage fork above, one branch has the insurer arguing supervisory failure[2]. Research establishing that supervisory failure is the predictable result of the arrangement does not help a practitioner establish that supervision was adequate. It tends the other way.

And in a professional standard setting, the argument that a failure was predictable is weak. A known and documented failure mode is one a competent professional is expected to have designed against, which is close to the technology competence framing above.

So the literature that exculpates the individual at the level of moral blame may aggravate their position at the level of professional and insurance responsibility, because it establishes that the risk was foreseeable.

We think the correct response is the one this series has recommended throughout, and it now has a second justification. Structural controls, being scheduled verification, constitutive review design, sampling of approvals and recorded reasons, are not merely better engineering. They are the evidence that the firm designed against a known failure mode rather than relying on a control the research says does not hold.

Documentation Does Double Duty

The synthesis that connects this article to the technical ones.

One source states it directly: documentation is the evidence of reasonable care, being the standard that determines both coverage eligibility and malpractice defence[2].

That single sentence reframes several recommendations made earlier in this series.

The incident response article argued for retaining prompt logs, model state, retrieval context and tool call history, on the grounds that they cannot be captured retroactively and that without them an investigation is reduced to inference. The lineage article argued for a correlation identifier so those records can be read together, and for recording ownership, meaning whose judgment shaped the output.

Both were presented as governance measures. On this source's framing they are also the file a practitioner would need to demonstrate what they actually did, in either forum.

The asymmetry is stark and worth stating plainly. A practitioner who can produce a contemporaneous record showing which sources were supplied, what the system produced, what the reviewer changed and on what basis is in a strong position. A practitioner whose record is a final deliverable and an approval timestamp is asserting that a review occurred and cannot demonstrate what it consisted of.

Since these records cannot be created after a claim arises, the decision about whether to keep them is made long before anyone knows whether they will be needed.

The Canadian Translation

What carries across and what does not, stated carefully because the differences are material.

The sources here describe an American commercial market for legal malpractice insurance. Canadian professional liability arrangements differ in ways that matter, including that several Canadian professions operate under mandatory insurance requirements or profession-sponsored programmes rather than purely open-market placement, and the specifics vary by profession and province.

We are not going to characterise any Canadian programme's terms, because we have not reviewed them and doing so accurately requires the actual wordings.

What carries across is the analysis rather than the answers. Any policy written before AI-assisted work existed will be interpreted against its existing language. The professional services versus technology failure characterisation is available to any insurer under any such wording. Renewal is the point at which terms can change. And a licensing body's jurisdiction is independent of any insurance arrangement.

So the questions a Canadian practitioner should put to their broker, insurer or programme administrator are these: is our cover silent or affirmative on AI-assisted work; would a verification failure be treated as a professional services claim; is there any technology or AI exclusion, and has one been added at any renewal; what is being asked about AI use, and what did we answer.

Every one of those is answerable, and the answers are specific to the arrangement rather than derivable from commentary about another country's market.

The Client Side

A parallel question this article should raise without pretending to answer it.

The preceding article addressed the vendor contract. There is a corresponding question on the client side: whether and how a firm should disclose AI use in its engagement terms.

Sources in this area discuss disclosure of AI use, warranty carve-outs for AI-generated portions, and liability caps as client-facing contract provisions[8], though the framework that source addresses is foreign.

Two considerations pull in different directions and both are real.

Disclosure supports informed consent and may be required or expected by a professional body, and it addresses the entitlement question raised in the previous article, being whether the firm is permitted to submit client information to a third-party processor at all.

Against that, a disclosure drafted as a disclaimer risks implying that AI-assisted work carries a lower standard, which is not a position a professional can take and which a professional body is unlikely to accept.

We would take that tension to counsel and to the relevant Canadian professional body's guidance rather than resolve it here, and we flag it because a firm that has addressed its vendor contract and not its engagement letter has done half the work.

A Worked Case: Two Characterisations

A Canadian firm whose AI-assisted deliverable contained an incorrect figure on which a client relied. The reconstruction illustrates the coverage analysis rather than reporting a specific matter.

The facts are agreed. The system extracted a figure incorrectly, the reviewer did not catch it, the deliverable issued, the client acted on it and suffered a loss.

On one characterisation this is a professional services claim: the practitioner exercised professional judgment in reviewing and adopting the work, and did so inadequately. That sits inside the insuring clause.

On another it is a technology failure: a tool produced a wrong output. That points toward technology exclusions or toward a cyber form that was not written for this[1][3].

The firm's position depends on facts it either recorded or did not. What was extracted, what source it came from, what the reviewer saw, what they changed, and on what basis. Without those it can assert that a review occurred and cannot show what it consisted of[2].

Separately, and regardless of how the claim resolves, the firm's professional body may take its own interest in whether the work met the standard[6], and no insurance outcome disposes of that.

Every variable that determined the outcome was fixed before the claim: the records kept, the design of the review, and the answers given at the last renewal.

What To Do

Build an inventory of where AI is actually in use. Deployed tools, features enabled inside existing software, and what staff use on their own initiative. You cannot answer the renewal question without it.

Ask whether your cover is silent or affirmative on AI-assisted work. Silence is ambiguity, not coverage, and the ambiguity is resolved by the party deciding whether to pay.

Get a written position on verification failures. Specifically, that they are treated as professional services claims rather than technology failures.

Read the renewal, not just the policy. Exclusions are reported to be arriving at renewal, which is a date already in your diary.

Treat the licensing exposure separately. A regulator's jurisdiction is independent of any policy, and no cover restores a licence.

Keep the records that show what the review consisted of. Sources supplied, output produced, changes made, basis recorded. They cannot be created after a claim.

Design against the documented failure modes. Foreseeability cuts against you, so structural controls are evidence of competence rather than merely good practice.

Take the engagement letter question to counsel. Vendor contract and client disclosure are two halves of the same problem.

Do not rely on the vendor agreement. As the preceding article set out, it is commonly drafted to cap the vendor's exposure at fees paid.

The Limits Of This Analysis

Several caveats matter and the first two are fundamental. Nothing here is legal advice or insurance advice; this publication is an accounting and advisory practice, and coverage questions are policy-specific and must be taken to a broker, insurer or coverage counsel. And the sources are overwhelmingly American and concern the legal profession, while Canadian professional liability arrangements differ materially, including through mandatory and profession-sponsored programmes whose terms we have not reviewed and do not characterise. Source quality varies: one discloses verification of carrier actions against primary or named broker sources, one is an insurance trade publication, several are commercial publications from a legal AI advisory site, and one is a general careers publication whose named-carrier claims we have not verified and which should be checked before reliance. The reported American conduct rules, the state adoption count and the malpractice survey findings are reported at second hand and unverified. Named insurers and products are reported without endorsement or verification. The inventory and renewal-answer argument, the observation that foreseeability may aggravate rather than mitigate a practitioner's position, the elaboration of regulatory exposure as structurally distinct from claims, the documentation synthesis and the worked case are our own analysis. This article does not address the terms of any Canadian professional programme, the rules of any Canadian professional body, quantum of cover, contractual limitation of liability with clients, or the position of unregulated advisers. Nothing here substitutes for advice from qualified counsel, a broker, or the reader's own professional body.

Related Reading

AI & Automation In Finance The Only Control That Binds The Vendor Why the vendor agreement will not carry the exposure this article describes. AI & Automation In Finance Your Policy Is Not The Variable Trust predicted whether employees hid their AI use more strongly than whether the organisation had an AI policy, and the finding held after controlling... AI & Automation In Finance Nothing Broke, And That Is The Problem The records that turn out to be the file you would need in either forum.

Frequently Asked Questions

Does using AI change the standard of care?

No. A professional who signs work is answerable for it, and the method of preparation has never been a defence. What changes is downstream: whether an insurer characterises the loss in a way the policy responds to, whether a regulator takes a separate interest, and whether the practitioner can demonstrate what their review consisted of.

Our policy does not mention AI. Are we covered?

Commentary describes that situation as creating ambiguity rather than coverage, with one carrier publicly calling AI claim coverage uncertain under current language. Coverage turns on whether the loss arose from professional services performed by the insured, and silence about a new method does not settle that question. The absence of the word is the problem, not the reassurance.

What is the recharacterisation risk?

That one set of facts supports two descriptions. A wrong figure can be a professional who exercised judgment badly, which sits inside a professional liability policy, or a tool that produced a bad output, which points toward technology exclusions or a cyber form. Sources describe a grey zone where neither policy clearly responds. Ask for a written position that verification failures are professional services claims.

Why does the renewal questionnaire matter?

Because it is where this arrives whether or not you have thought about it, and because answering accurately is harder than it sounds. A firm may have no deployed AI tool while running AI features inside its accounting platform and document system, with staff using assistants independently. A firm that cannot enumerate its AI use is not in a position to answer the question.

Does insurance address our regulatory exposure?

No, and this is the point most relevant to licensed professionals. A licensing body can sanction a member regardless of who is sued. Discipline is a proceeding about standards, not a claim for money, and the two are independent: discipline can follow with no claim, and a claim can settle within cover while a separate process continues. No policy restores a licence.

Does the research on automation bias help our defence?

Probably not, and possibly the reverse. Research establishing that reviewer complacency is predictable and cannot be trained away does not show that supervision was adequate, and a documented, foreseeable failure mode is one a competent professional is expected to have designed against. That makes structural controls evidence of competence rather than merely good engineering.

References

1. AI Vortex. (2026, April 11). AI Malpractice Insurance for Law Firms: What Exists in 2026, on most pre-AI policies creating ambiguity rather than coverage, technology failure exclusions being invoked to deny or limit cover, the grey zone where neither professional liability nor cyber clearly responds, named carriers developing questionnaires and endorsements, exclusions appearing in new renewals at an accelerating rate, and a reported malpractice survey finding AI-related claims among the fastest-growing categories with most pre-2023 policies containing no AI language. Note: a commercial legal-AI advisory publication; the survey and carrier positions are reported at second hand. aivortex.io/legal/ai-governance/ai-insurance-liability-law-firms
2. AI Vortex. (2026, April 17). AI Malpractice Insurance Coverage, on policies covering wrongful acts in the performance of professional services and the fork between arguing the AI performed the work and arguing supervisory failure, a carrier publicly describing AI claim coverage as uncertain under current language with that uncertainty benefiting the insurer, exclusions that could defeat a claim, and documentation as the evidence of reasonable care determining both coverage eligibility and defence. Note: a commercial publication that discloses the article was researched and written with AI assistance and reviewed by a named editor. aivortex.io/legal/ai-governance/ai-malpractice-insurance-coverage
3. AI Vortex. (2026, April 17). The AI Liability Gap Law Firms Aren't Covering, on the scenario in which an insurer recharacterises a claim as a technology failure rather than a professional judgment error to avoid coverage, the recommendation to secure an endorsement explicitly covering AI-assisted work product, ensuring policy language does not exclude technology-assisted services, and obtaining written confirmation that AI verification failures are covered as professional services claims. Note: a commercial publication. aivortex.io/legal/ai-governance/ai-professional-liability-gap
4. Legal AI Governance. (2026, June 19). AI Liability Insurance for Law Firms: A Primary-Source Guide, on the distinction between silent AI cover, where an unamended form is adjudicated under pre-existing language, and affirmative AI cover; the observation that whether the policy pays is the load-bearing question and that the 2026 answer is bifurcated; and a quoted broker risk-control lead for accounting and law firm professional services describing carriers asking firms whether they use AI. Note: the source discloses that carrier actions and form citations are verified against primary sources or named broker and trade-press confirmation, and states it is not legal or insurance advice. legalaigovernance.com/resources/ai-liability-insurance
5. Insurance Insider. (2026, May 13). Lawyers Test Case for Professional Liability AI Risks, on errors and omissions policies covering legal fees, settlements and damages for professionals including accountants; one insurer leading on broad AI exclusions in professional liability forms since 2024; and the trend in the legal space as a possible precursor to wider AI exclusions across the professional lines market. Note: an insurance trade publication; accessed in summary. insuranceinsider.com
6. Metaintro. (2026, May 16). Insurers Are Refusing to Cover AI Mistakes, on licensed professionals in medicine, law and accounting facing an additional layer because a licensing body can sanction them for AI-assisted mistakes regardless of who is sued, and on reported regulatory filings and approvals by several named major carriers to exclude AI-related claims from commercial liability and directors and officers coverage. Note: a general careers publication rather than an insurance trade source; the named-carrier filing claims were not verified and should be checked before reliance. metaintro.com/blog/insurers-refusing-cover-ai-mistakes-2026-job-impact
7. AI Legal Authority. AI Use and Legal Malpractice Risk: Standards of Care and Insurance Implications, on exposure attaching differently depending on which stage AI touches, how much supervision occurred and whether output was verified before reliance; and on the American conduct baseline of competence, diligence and supervision, with a comment stating competence includes the benefits and risks associated with relevant technology, and a reported count of state adoptions. Note: a commercial publication describing a foreign regime; the state adoption count was not verified. ailegalauthority.com/ai-legal-malpractice-risk
8. Agent Mode AI. (2026, May 12). AI Client Contract Clauses, on client-facing provisions including disclosure of AI use, an intellectual property warranty carve-out for AI-generated portions, exclusion of client materials from training, and a liability cap tied to fee paid. Note: a commercial publication addressing a foreign regulatory framework; cited here only for the categories of client-side provision it identifies. agentmodeai.com/operators/ai-client-deliverable-contract-clauses

This article discusses professional liability and insurance and is provided for general informational purposes only. It is not legal advice or insurance advice. Sources are overwhelmingly American and concern the legal profession; Canadian professional liability arrangements differ materially and are not characterised here. Coverage questions are policy-specific and should be taken to a broker, insurer or coverage counsel, and conduct questions to the reader's professional body.